Model Card for cardio-ner-it-cardioberta-multilabel

This is a DT4H/CardioBERTa.it base model finetuned for span classification on Italian cardiology clinical text, using IOB tagging.

Expected input and output

The input should be a string of Italian cardiology clinical text.

cardio-ner-it-cardioberta-multilabel is a multilabel span classification model (labels are independent per token, so overlapping entity types are possible). The classes that can be predicted are: DISEASE, MEDICATION, PROCEDURE, SYMPTOM.

Extracting span classification from cardio-ner-it-cardioberta-multilabel

from transformers import pipeline

ner_pipe = pipeline('ner',
                     model="cardio-ner-it-cardioberta-multilabel",
                     tokenizer="cardio-ner-it-cardioberta-multilabel",
                     aggregation_strategy="simple",
                     trust_remote_code=True)

named_ents = ner_pipe(SOME_TEXT)

For text longer than the model's max sequence length, use a sliding stride:

named_ents = ner_pipe(SOME_TEXT, stride=125)

Data description

10-fold cross-validation on the Italian portion of CardioCCC (DataTools4Heart Cardiology Clinical Case Corpus), batches 1+2, 508 documents, version 1_validated_without_sugs. The uploaded checkpoint is the arithmetic mean of the 10 per-fold checkpoints ("Run 1"/weight-averaging in the CardioLM paper).

Acknowledgement

This is part of the DT4H project.

For more details about training/eval and other scripts, see the CardioNER github repo and for more information on the background, see DataTools4Heart's Huggingface/Website.